artemka081

- Developed a predictive model for railway infrastructure failures (F1 = 0.84, critical incident recall = 0.91) with high interpretability using SHAP, LIME, and Feature Importance. - Implemented the Talk2Model system to explain predictions in natural language. - Replaced the Word2Vec model with RoBERTa, improving clustering accuracy in an employee churn prediction project. - Accelerated ML models by introducing a feature selection system based on NSGA-II and bio-inspired algorithms. - Designed an anomaly detection system for IT monitoring data to predict service failures. - Initiated the integration of MLflow to ensure logging and reproducibility of team experiments. - Conducted onboarding and mentorship for interns, training them in ML fundamentals and model interpretability. Stack: Python, CatBoost, XGBoost, Scikit-Learn, PyTorch, DEAP, Pymoo, SHAP, LIME, Alibi Explain, MLflow, ClickHouse, Docker, GitLab

- Developed backends for web applications using Python Django, FastAPI, and Flask. - Created a Telegram bot with AIOgram, leveraging LLM and LangChain for analyzing RSS news feeds, generating recommendations for authors, and selecting relevant content. - Trained an LSTM neural network to detect text generated by LLMs (F1 = 0.62). - Implemented a RAG system for managing a personal knowledge base using LLama3, Chroma, and LangChain. - Conducted research on applying multi-objective optimization algorithms to create an ideal cryptocurrency portfolio. Stack: Python, Django, FastAPI, RabbitMQ, Keras, LangChain, Chroma, AIOgram, PostgreSQL, SQLAlchemy, Docker, GitHub